AI summaries can downplay medical issues for female patients, UK research finds

The article discusses a new study that found AI language models, specifically Meta's Llama 3 and Google's Gemma, were more likely to omit important medical information, such as "disabled," "unable," or "complex," when summarizing case notes for female patients compared to male patients. The research, led by the London School of Economics and Political Science, examined 617 adult social care worker case notes and found that the AI tools often provided very different patient snapshots depending on the gender of the patient. The article also highlights the broader issue of bias in medical AI, which can lead to women, racial and ethnic minorities, and the LGBTQ community receiving insufficient or inaccurate medical care. The article emphasizes the need for transparency and careful oversight in the deployment of these AI models, as they are being widely used in care practices without always detailing which models are being introduced or in what capacity.